Distribued estimation architectures and algorithms for formation flying spacecraft

Abstract

Future formation flying missions are being planned for fleets of spacecraft in MEO, GEO, and beyond where relative navigation using GPS will either be impossible or insufficient. To perform fleet estimation for these scenarios, local ranging devices on each vehicle are being considered to replace or augment the available GPS measurements. These estimation techniques need to be reliable, scalable, and robust. However, there are many challenges to implementing these estimation tasks. Previous research has shown that centralized architecture is not scalable, because the computational load increases much faster than the size of the fleet. On the other hand, decentralized architecture has exhibited synchronization problems, which may degrade its scalability. Hierarchic architectures were also created to address these problems. This thesis will compare centralized, decentralized, and hierarchic architectures against the metrics of accuracy, computational load, communication load, and synchronization. It will also briefly observe the performance of these architectures when there are communication delays.

(cont.) It will examine the divergence issue with the EKF when this estimator is applied to a system with poor initial knowledge and with non-linear measurements with large differences in measurement noises. It will analyze different decentralized algorithms and identify the Schmidt-Kalman filter as the optimal algorithmic choice for decentralized architectures. It will also examine the measurement bias problem in the SPHERES project and provide an explanation for why proposed methods of solving the bias problem cannot succeed. Finally, the SPHERES beacon position calibration technique will be proposed as an effective way to make the SPHERES system more flexible to a change of testing environment.